SkarpChapter 14 of 17
Legal Sourcework and the Eight Grand Challenges
The references reveal the dense legal and policy environment in which the proposal sits, while Annex I turns strategic ambition into eight defined technology missions. Together they show both where the framework comes from and which capabilities it aims to produce.
1. Reading the Text: Sources, Status, and Scope
A proposal, not an adopted Regulation
The text is part of COM(2026) 502 final, dated 3 June 2026. As of 20 July 2026, it remains a Commission proposal in an ongoing ordinary legislative procedure.
Two layers of the source
The references map the legal and policy environment. ANNEX I then converts that setting into eight proposed technology missions called Grand Challenges.
Read verbs carefully
Targets, examples, potential applications, and statements about what programmes "could" or initiatives "may" do have different force. Do not turn every aspiration into a binding duty.
2. The Reference List as Legal Sourcework
Different kinds of sources
The reference list combines binding Regulations and Directives, funding programmes, Commission communications, Council conclusions, recommendations, and reports. Their legal roles are not identical.
Core regulatory environment
The cited framework includes the Data Act, NIS 2 Directive, AI Act, Cyber Resilience Act, Interoperable Europe Act, energy-efficiency rules, and the Gigabit Infrastructure Act.
Do not infer missing details
References 7 and 8 point to SME and small mid-cap definitions, but this excerpt does not reproduce the criteria. The correct sourcework move is to consult the cited instrument.
3. Placeholders, Budget Notes, and Drafting Discipline
A placeholder is not a legal citation
Reference 14 says: "References to Chips Act 2.0 to be added once adopted." It does not establish that an adopted Chips Act 2.0 is already cited by this text.
Two more unfinished references
The source also says: "Add reference to the Regulation on speeding-up environmental assessments once adopted." It likewise points to a proposed European Competitiveness Fund.
Financial precision
"Diff. = Differentiated appropriations / Non-diff. = Non-differentiated appropriations." The source also says: "Budget lines for the new MFF are not yet known".
4. Grand Challenge 1: Sustainable, High-Performing, Secure Data Centres
The PUE target
Grand Challenge 1 targets "an average Power Usage Effectiveness (PUE) of 1.15 across the Union." It links the target to cloud and edge data-centre sustainability and performance.
The utilisation direction
It also calls for "raising average server utilisation rates across the Union’s data centres towards 50%", including AI-powered management, scheduling, and balancing of operational constraints.
Security is part of the mission
The challenge also addresses supply-chain resilience, Union-designed and manufactured semiconductor and quantum technologies, and resistance to physical and cybersecurity threats.
5. Grand Challenges 2 and 3: Cloud Capacity and Frontier AI
Challenge 2: the full stack
Challenge 2 is "Building end-to-end hardware and software cloud stacks, including AI tools, infrastructure, services and management layers" to bridge critical capacity gaps.
Distributed AI infrastructure
The challenge includes AI servers using Union-designed and manufactured semiconductors and quantum technologies for distributed and decentralised cloud and edge computing for AI.
Challenge 3: capability frontier
Frontier AI focuses on multimodal models, advanced reasoning, cross-modal understanding, agentic capabilities, efficiency, cognitive modelling, and alternative computational structures.
6. Grand Challenges 4, 5, and 6: AI in the Physical and Industrial World
Physical AI
Challenge 4 combines software-hardware co-design, frontier AI, and world models so systems can manipulate, navigate, and interact robustly in unstructured environments with minimal supervision.
Industrial AI validation
Challenge 5 expects specialised computing resources and testing facilities to validate AI in real-world environments before large-scale deployment and uptake.
Cooperation without disclosure
Challenge 6 develops industrial systems "without exposing commercially sensitive data between participants." Its mechanisms include federated training, encryption, and secure execution.
7. Grand Challenges 7 and 8: Coordinated Agents and Public Services
Challenge 7: orchestration
The AI Agents Platform is a proposed European orchestration framework: middleware for resilient, secure deployment and large-scale management of autonomous AI agents.
More than standalone agents
The challenge explores collaboration among multiple agents while maintaining rigorous security, and calls for resilient, cloud-based open platforms.
Challenge 8: critical public services
Public Sector AI is "Developing AI models and systems, based on high-quality data from the public sector targeting critical domains" including healthcare, law, crisis management, and administration.
8. Flashcards: Precision Recall
Flip each card and test whether you can connect the exact term or phrase to its function in the source.
- PUE target
- Grand Challenge 1 seeks "an average Power Usage Effectiveness (PUE) of 1.15 across the Union."
- Server-utilisation direction
- Grand Challenge 1 calls for "raising average server utilisation rates across the Union’s data centres towards 50%".
- Cloud stacks
- "Building end-to-end hardware and software cloud stacks, including AI tools, infrastructure, services and management layers" is Grand Challenge 2.
- Cooperative model safeguard
- Grand Challenge 6 enables industrial-scale collaboration "without exposing commercially sensitive data between participants."
- Public Sector AI
- "Developing AI models and systems, based on high-quality data from the public sector targeting critical domains" is Grand Challenge 8.
- Differentiated appropriations
- "Diff. = Differentiated appropriations / Non-diff. = Non-differentiated appropriations."
- MFF budget-line limitation
- "Budget lines for the new MFF are not yet known" means this excerpt does not allocate a named future budget line to a Grand Challenge.
9. Quiz: Targets and Legal Reading
Choose the answer that most precisely reflects Annex I and the document's current status.
Which statement is most accurate?
- Grand Challenge 1 sets an Annex I target of an average PUE of 1.15 across the Union and directs average server utilisation towards 50%.
- Grand Challenge 1 requires every individual data centre to achieve exactly PUE 1.15 and exactly 50% utilisation immediately.
- Grand Challenge 2 sets the PUE target, while Grand Challenge 3 sets the server-utilisation target.
- Because COM(2026) 502 final is a proposal, Annex I contains no stated targets at all.
Show Answer
Answer: A) Grand Challenge 1 sets an Annex I target of an average PUE of 1.15 across the Union and directs average server utilisation towards 50%.
Option 1 closely follows the source. Grand Challenge 1 states an average Union-wide PUE target of 1.15 and seeks to raise average server utilisation towards 50%. The Annex does not say every individual facility must reach those exact figures immediately. The proposal status does not erase the targets; it affects whether they are already binding law.
10. Quiz: Matching a Mission to Its Safeguard
Identify the Grand Challenge whose design directly addresses the stated information-sharing constraint.
A group of aerospace firms wants to train a shared AI system but cannot centrally disclose commercially sensitive design data. Which Annex I mission most directly fits this situation?
- Grand Challenge 2: Cloud stacks, because it is mainly about end-to-end cloud layers.
- Grand Challenge 4: Physical AI, because it is mainly about robots in unstructured environments.
- Grand Challenge 6: Cooperative European Industrial Models, because it supports collaboration without exposing commercially sensitive data between participants.
- Grand Challenge 8: Public Sector AI, because all sensitive data are public-sector data.
Show Answer
Answer: C) Grand Challenge 6: Cooperative European Industrial Models, because it supports collaboration without exposing commercially sensitive data between participants.
Grand Challenge 6 is the direct match. It explicitly aims to enable collaboration at European industrial scale "without exposing commercially sensitive data between participants" and lists federated or distributed training, secure execution environments, encryption-based processing, anonymisation, pseudonymisation, access compartmentalisation, and anti-extraction protections.
Key Terms
- MFF
- Multiannual Financial Framework. The source states: "Budget lines for the new MFF are not yet known".
- PUE
- Power Usage Effectiveness, used in Grand Challenge 1 as the metric for the stated average target of 1.15 across the Union.
- Annex I
- The part of the proposed Cloud and AI Development Act that sets out the eight Grand Challenges.
- Middleware
- A coordination layer; Grand Challenge 7 describes an AI agent orchestration framework as essential middleware for deploying autonomous agents at scale.
- Cloud stack
- The connected hardware and software layers that include AI tools, infrastructure, services, and management layers.
- Physical AI
- AI systems designed to operate autonomously and safely in physical, dynamic, and unstructured environments.
- Synthetic data
- Artificially generated data; the source identifies high-fidelity synthetic data generation as a privacy-preserving framework for Public Sector AI.
- Federated learning
- A confidentiality-preserving approach identified in the source in which algorithms are brought to data rather than centrally transferring the data.
- Server utilisation
- The extent to which server capacity is actively used; Grand Challenge 1 seeks to raise the Union-wide average towards 50%.
- Multimodal frontier AI
- Next-generation AI models and systems intended to advance reasoning, cross-modal understanding, and agentic capabilities.
- Differentiated appropriations
- Appropriations defined in the source through the exact note: "Diff. = Differentiated appropriations / Non-diff. = Non-differentiated appropriations."